The model, at least in its present form, is judged to be unsatisfactory; some general conclusions will be drawn at the end of the paper (section 8 below). READ stands for "Regional Energy Activity and Demography". large-scale annual econometric model of the United States, with very fine detail. The basic geographical unit is the county, and there are almost fifty industries in the present version of the model. The object is to analyze the impact of energy policy on regional economic activity. The model has four basic sectors: industrial location (output by industry and county) demography (employment by industry and county, labor force and population by county) construction (over a hundred building types, by industry and county) Energy prices appear as explanatory variables in most of the econometric equations. There is also a linear-programming transportation sub-model, which in effect moves industrial output over a transportation network with almost five hundred nodes. Shadow prices from this sub-model are used as explanatory variables in the industrial location sector. Macro-economic variables (GNP and its major components) are treated as exogenous, and can be supplied to READ from any of the standard macro-models. Regional energy prices too are exogenous, and will be supplied to READ from in-house EIA (Energy Information Administration) models like PIES and its sucIn effect, READ disaggregates the macro-level forecasts down to the Section 1. Introduction county level. Then, it is possible to aggregate back up to any desired geographical level. The fitting period for the model is 1965-74, and forecasts will be made out to 1990. READ is based on the Maryland model of Harris and Hopkins [1], with further development work by Hopkins and others, in the Office of Applied Analysis Energy Information Administration After the initial development phase, but before the equations were fitted, EIA decided to review the model. The review process will be discussed in section 2 of this paper. Section 3 will briefly describe the industrial location and demographic sectors of the model. Section 4 presents a critique of the equations, and section 5 a critique of the fitting procedures. The data problems are considered in sections 6 and 7. Conclusions will be found in section 8. 2. The READ review The review of the READ model was organized by George Lady Office of Applied Analysis Energy Information Administration Washington, D.C. and by two of his consultants, Daniel Khazzoom and Richard Ruppert. There were seven academic reviewers: David Brillinger, U.C. Berkeley Leon Cooper, Southern Methodist University Robert Dorfman, Harvard David Freedman, U.C. Berkeley Jerry Hausman, M.I.T. Karen Polenske, M.I.T. Harvey Wagner, University of North Carolina There was also a meta-reviewer, David Wood (M.I.T.): his function was to review the review process. The reviewers were given nearly a thousand pages of documentation on the READ model, prepared by the modelling group. Then, a meeting was held at which the model was described and discussed. On the basis of this discussion, George Lady drew up a list of questions, which were answered in writing by the reviewers. The READ Model A second meeting was held to discuss the answers. It became clear that the model had very serious data problems. Indeed, a consensus was reached that as specified, and based upon currently available data, the READ model does not justify the resource commitment necessary for its continued development [2]. The issue was then forced: how should EIA do regional impact analysis? A third review meeting was held to consider this question. Three main options were considered: Revising READ, for example by increasing the level of geographical No consensus was reached. The reviewers did agree on the following point: In a large econometric model with the same fitting period as READ, it would be very difficult to demonstrate the impact of energy prices on industrial location of demographics. The wisdom of developing such a model to do regional impact analysis is therefore questionable. The main characteristics of the READ review can be summarized as follows: • The process was reasoned argument, based mainly on documentation • The reviewers drew on their knowledge of related fields like econometrics, linear programming and statistics. No attempt was made to compare model forecasts to actual observations, for the following reasons: • The equations had not yet been fitted, so the model was not in a position to make point forecasts. • No explicit strategy was available for forecasting exogenous variables. Section 2. The READ review • There was no valid procedure for estimating the coefficients, or measuring the uncertainty in model forecasts of endogenous variables (section 5 below) • Most of the endogenous variables in the model are not measured, so there was nothing against which to compare forecasts (section 6 below). 3. A closer look at the READ model This section will describe the following components of the READ model: "Industry" is a catch-all term. READ industry #1 is "agricultural production," and READ industry #40 is "wholesale and retail trade." The READ industries are defined in terms of the usual Standard Industrial Classification or SIC codes [5]. For instances, READ industry #1 corresponds to SIC codes 01-02, and READ industry #40 corresponds to SIC codes 50-59. The present version of the model has 47 industries, a private household sector, and several government sectors (local, state and federal). "Industrial location" is a bit of a misnomer. This sector predicts annual changes in the value of output (e.g., sales) by READ industry and county, using a linear regression equation. A "change" is the difference between the value of output for the current year and for the previous year, both measured in 1967 dollars. The independent variables are the following: energy prices transportation costs (shadow prices from the transportation sub-model described below) |